Phenology-driven multisensor unsupervised crop mapping in data-scarce irrigated agriculture: a case of Adea District in Ethiopia

Accurate irrigation planning, water allocation, and agricultural productivity assessment are frequently constrained in data-scare regions where reliable ground-based crop information and spatial coverage data are limited. This study addresses these gaps using a phenology-driven, multisensor unsupervised crop-mapping framework implemented within Google Earth Engine (GEE). The approach integrates monthly normalized difference vegetation index (NDVI) profiles from Sentinel-2, Sentinel-1VV backscatter, and digital elevation model–derived terrain slope into unsupervised K-means clustering algorithm. Adea District, Ethiopia, served as a demonstration site to evaluate and interpret the phenological signatures of key irrigated crops. The identified crop types were mapped to estimate spatial coverage and compared with the ground truth observations of irrigated wheat and tomato fields. The approach successfully distinguished irrigated wheat and tomato fields with an overall classification accuracy of 91%, a Kappa coefficient of 0.82, and estimated cultivated areas of 666.93 and 812.35 ha, respectively. Combining phenological profiles with multisensor satellite observations significantly enhance unsupervised classification performance in heterogeneous, smallholder-dominated agricultural landscapes. The cloud-based framework offers a computationally scalable and cost-effective solution for crop mapping, and water productivity monitoring in data-scarce agricultural regions.